AI 资源精选
  首页 >大语言模型 >LLaVA++
访问数:254
LLaVA++

厂家:ORYX

简介:

LLaVA++项目通过扩展现有的LLaVA模型,成功地为Phi-3和Llama-3模型赋予了视觉能力。这一改进标志着AI在多模态交互领域的进一步发展。

  详细介绍 - LLaVA++

性能雷达

下图是有关LlaVa-7B,13B,Phi3-V -3.8B和LLaMA3-V-8B的性能雷达图,供参考:


lava++_radar_plot.png


模型公园

更多模型下载地址:(如果无法下载,请将链接中的huggingface.co替换为hf-mirror.com即可用镜像下载)

基于Phi-3的模型:

Model NameHugging Face LinkSummary
LLaVA-Phi-3-mini-4k-instruct-pretrainHugging FacePretrained on LCS-558K.
LLaVA-Phi-3-mini-4k-instruct-loraHugging FaceLoRA weights fine-tuned on LLaVA-Instruct-665K.
LLaVA-Phi-3-mini-4k-instructHugging FaceMerged LoRA weights in HuggingFace format.
LLaVA-Phi-3-mini-4k-instruct-FTHugging FaceFully fine-tuned model weights in HuggingFace format.

基于Llama-3的模型:

Model NameHugging Face LinkSummary
LLaVA-Meta-Llama-3-8B-Instruct-pretrainHugging FacePretrained on LCS-558K.
LLaVA-Meta-Llama-3-8B-Instruct-loraHugging FaceLoRA weights fine-tuned on LLaVA-Instruct-665K.
LLaVA-Meta-Llama-3-8B-InstructHugging FaceMerged weights in HuggingFace format.
LLaVA-Meta-Llama-3-8B-Instruct-FTHugging FaceFully fine-tuned model weights in HuggingFace format.
LLaVA-Meta-Llama-3-8B-Instruct-FT-S2Hugging FaceFully fine-tuned S2 model weights in HuggingFace format.


--文 by AixTong.com--
相关资讯 / AI资源